🌟Samsung Officially Releases Galaxy S25 Series
📸Instagram Offers Big Bonuses to Attract TikTok Creators
👓Meta Reportedly Developing Wearable Hardware to Compete with Apple
🔍Google Invests Over $1 Billion in Anthropic
🚗Volkswagen to Release No New Electric Vehicles in 2025
📈ByteDance’s Doubao Model Achieves 50% Gross Margin After Price Cut
🇨🇳Nothing to Enter Chinese Market Soon
💡Mirumi Creator: The More Human-like Robots Are, the Harder They Are to Accept
📱iPhone 17 Series Prototype Shells Allegedly Leaked
📜NetEase Youdao Officially Releases Ziyue -o1 Reasoning Model
🎉Doubao Model 1.5Pro Officially Released
Samsung Officially Releases Galaxy S25 Series
Samsung held its global Galaxy product launch event on Jan 23.
The Samsung Galaxy S25 series includes three models: S25, S25+, and S25 Ultra. Additionally, the long-rumored super slim Galaxy S25 Edge was also unveiled.
In terms of appearance, all three new models in the Samsung Galaxy S25 series share a consistent design language, with 14 color options available in the Chinese market (including Samsung Store exclusive colors). Compared to the Galaxy S24 series, the S25 series features more rounded corners for a more comfortable grip. The S25 Ultra adopts the same streamlined design as the S25, moving away from the previous generation’s more angular style.
The Samsung Galaxy S25 Ultra features a titanium alloy frame and weighs 218g, 15g lighter than the previous generation. The Galaxy S25 and S25+ use a recycled armor aluminum frame for the first time, with the lightest Galaxy S25 weighing only 162g.
Performance-wise, all three new models are equipped with the Qualcomm Snapdragon 8 Supreme Edition processor customized for Samsung, offering nearly 40% performance improvement over the previous generation. They also feature a new cooling architecture to ensure performance release. Combined with the new One UI 7 system, they offer more AI functions, such as an AI assistant that can execute tasks across applications with just one voice command.
In terms of imaging, all three new models are equipped with a 200MP wide-angle main camera, a 10MP 3x telephoto lens, and a 50MP ultra-wide-angle lens. The Galaxy S25 Ultra also has a 50MP 5x telephoto lens. For video recording, the Galaxy S25 series supports 10-bit HDR video log recording and has improved low-light video capabilities.
All three new models start with 12GB+256GB configurations. The international and Chinese “early bird” prices for the Samsung Galaxy S25 series are as follows:
- Galaxy S25 starts at $799.
- Galaxy S25+ starts at $999.
- Galaxy S25 Ultra starts at $1,299.
Additionally, the highly anticipated Samsung Galaxy S25 Edge was also unveiled at this event. Its design language is consistent with the officially released Galaxy S25 series, featuring a rectangular frame with a rumored thickness of around 6.5mm. Notably, despite its slim body, the Galaxy S25 Edge still includes a dual rear camera module.
The official details about this super slim phone are still scarce, but it is rumored to be positioned between the Galaxy S25+ and S25 Ultra, with a likely price of $999.
Instagram Offers Big Bonuses to Attract TikTok Creators
Recently, The Information reported that Meta’s Instagram is offering monthly bonuses ranging from $10,000 to $50,000 to attract video creators to post short videos on Instagram Reels. Creators must first publish on Instagram Reels before sharing on other platforms.
Despite TikTok resuming services in the U.S., it has not returned to the Apple App Store or Google Play Store, presenting an opportunity for Instagram.
Meta spokesperson Paige Cohen told The Verge about a “breakthrough bonus” program encouraging TikTok creators to earn up to $5,000 by posting Reels on Instagram and Facebook over three months. Cohen also mentioned that Meta plans to offer content collaboration opportunities to TikTok creators in the coming months to help them expand their influence on Instagram and Facebook.
Recently, Instagram CEO Adam Mosseri announced a new mobile video editing app called “Edits.” Notably, on the same day, ByteDance’s “CapCut” (international version) ceased operations in the U.S. due to a ban.
Mosseri stated that “Edits” aims to provide creators with the best possible tools, regardless of future developments.
Meta Developing Wearable Hardware to Compete with Apple
On January 22, Bloomberg’s Mark Gurman reported that Meta is upgrading its smart glasses and exploring new wearable devices.
The report indicates that Meta plans to develop products to rival Apple’s Apple Watch and AirPods. About five years ago, Meta explored launching a smartwatch to compete with Apple and Samsung, but the project was repeatedly canceled. Now, Meta intends to revive the smartwatch project, using the device to display photos taken by smart glasses.
Additionally, Meta is developing headphones to compete with AirPods, featuring a camera to capture external information and enhance user experience through AI analysis. Internally known as “Camera Buds,” the product is still in early stages.
Insiders revealed that Meta plans to collaborate with eyewear brand Oakley in 2025 to develop smart glasses for athletes based on Oakley’s Sphaera model, codenamed “Supernova 2,” with a camera in the center of the frame.
Meta’s hardware subsidiary, Reality Labs, plans to release high-end glasses with built-in displays in 2025, codenamed “Hypernova,” resembling Ray-Ban glasses. The right lens will have a display to show information to users, and simple apps can run on the glasses. Employees involved in the project estimate Hypernova will cost around $1,000.
In September 2024, Meta announced the world’s most expensive AR glasses, Meta Orion, with a 10-year development time and a cost of $10,000. However, Gurman reported that Orion glasses will not be sold to customers; instead, Meta aims to launch a successor, codenamed “Artemis,” by 2027. Sources familiar with the Artemis prototype say it is more advanced and lighter than the Orion test version.
Google Invests Over $1 Billion More in Anthropic
On January 22, the Financial Times reported that Google is investing over $1 billion more in Anthropic, a competitor to OpenAI.
The report states that Google had previously invested about $2 billion in Anthropic, and insiders reveal that Google is increasing its stake. Notably, Amazon has invested a total of $8 billion in Anthropic over the past 18 months and is working to integrate Anthropic’s Claude model into the next-generation Alexa AI speaker.
Anthropic was founded in 2021 by siblings Dario and Daniela Amodei, both former OpenAI executives. The Claude model developed by Anthropic is considered a major competitor to OpenAI’s GPT-4 model.
Volkswagen to Release No New Electric Vehicles in 2025
On January 22, Carscoops reported that Volkswagen has no plans to release new electric vehicles in 2025, and the highly anticipated ID.2 model will not be available until 2026.
The report reveals that Volkswagen does not plan to launch any new electric models in 2025. The latest model, the ID.7, was released in Q3 2023, meaning no new models will be released in 2024 either. German automotive analyst Matthias Schmidt suggests that Volkswagen may face significant challenges in 2025.
The ID.2 model, expected to be priced under €25,000, will offer interior space similar to the Volkswagen Golf while maintaining the compact body and affordability of the Volkswagen Polo.
The report also mentions that the first model developed in collaboration with Chinese automaker XPeng will be launched in 2026. In February 2024, XPeng and Volkswagen signed a joint development agreement for platform and software strategic technology. In July of the same year, they signed an agreement for electronic and electrical architecture technology, with the first model featuring the jointly developed architecture expected to be mass-produced within 24 months.
The first models developed by Volkswagen and XPeng will include two vehicles, one of which is an SUV, primarily targeting the Chinese electric vehicle market. Volkswagen aims to increase its annual sales in China from 2.93 million units in 2024 to 4 million units by 2030.
ByteDance’s Doubao Model Achieves 50% Gross Margin After Price Cut
On January 22, Jiemian News learned from multiple insiders that the Doubao model’s gross margin remains positive after a significant price cut in 2024.
The report states that ByteDance’s latest Doubao model 1.5 has achieved further breakthroughs in inference cost optimization. The Doubao-1.5-pro, sold on the Volcano Engine API, still maintains a 50% gross margin.
In May 2024, ByteDance officially released the Doubao model. The pro-32k version of the general model has an inference input price of only about $0.11 per million tokens, while similar models on the market were priced at $16.42 per million tokens, 150 times the price of the Doubao model. After the Doubao model announced a price cut, other products like Chovm Cloud’s Tongyi model also lowered their prices.
There are doubts in the market about whether the model can still be profitable, but insiders say these concerns seem unfounded. Volcano Engine President Tan Dai previously stated in an interview that a large enough usage volume is needed to refine a good model and significantly reduce the unit cost of model inference. However, insiders reveal that due to significant R&D investment, ByteDance’s model business is still operating at a loss.
As of now, ByteDance has not confirmed the above information.
Nothing Decided to Enter the Chinese Market
On January 21, a Xiaohongshu user named Yan Lin, described as the “Industrial Design Director of Nothing’s Phone Business,” announced that “Nothing is officially entering the Chinese market.”
Recently, the Nothing team held its first media communication meeting in China. At the event, Nothing showcased its entire product lineup, including phones, watches, various earphones, and a desktop charger that was developed in the lab but ultimately not released. The charger supports wireless charging and 2C1A wired charging.
At the communication meeting, Nothing directly addressed the question of whether its phones would enter the Chinese market, stating that there are currently no plans for this.
Nothing was founded in 2020 by former OnePlus co-founder Carl Pei and has launched several products, including the Nothing Ear series and Nothing Phone series. It also launched the CMF sub-brand, focusing on modular design digital products.
Mirumi Creator: The More Human-like a Robot, the Harder It Is to Accept
On January 20, Geek Park published a conversation with Shunsuke Aoki, founder of Japanese robotics company Yukai Engineering. The company recently showcased a companion robot named “Mirumi” at CES.
Shunsuke Aoki stated that Mirumi’s design was inspired by babies, but he ultimately did not design Mirumi to look like a baby. He believes, “If it looks too human, we have to treat it as a person. That’s why we avoid making robots look too human. We want people to feel joy and comfort from robots, not discomfort.”
Regarding the absence of AI in Mirumi, Shunsuke Aoki explained, “We’re unsure about adding AI features.”
He mentioned that their philosophy is to start with simple products and gradually add features, similar to software development. If people were given a complex app like WeChat from the start, they would be confused. While AI features can enhance interaction, using them requires a monthly fee, which is a high demand from a business perspective.
NetEase Youdao Officially Launches Ziyue-O1 Reasoning Model
On January 22, NetEase Youdao announced the release of Ziyue-O1, China’s first reasoning model capable of providing step-by-step explanations. The model has also been open-sourced simultaneously.
According to the introduction, as a lightweight single model with 14B parameters, Ziyue-o1 supports deployment on consumer-grade graphics cards. It uses chain-of-thought technology to provide detailed problem-solving processes, achieving higher accuracy in problem-solving with strong logic and reasoning capabilities, and offers Chinese logical reasoning.
While “compressing” the scale, Ziyue-o1 uses chain-of-thought technology to create China’s first chain-of-thought model that outputs step-by-step explanations. With a small parameter scale of 14B, it can reproduce the single-model reasoning capabilities of OpenAI o1.
Ziyue-o1 has been applied in “Youdao Xiao P,” supporting its Q&A process of “providing the solution idea first, then the answer,” guiding student users to think actively and use their knowledge reserves to solve problems independently.
Youdao Ziyue-o1 has provided a demo version and is simultaneously available on Huggingface and ModelScope.
Doubao Large Model 1.5Pro Officially Released
On January 22, 2024, Doubao officially announced the launch of the Doubao Large Model 1.5Pro version.
It is understood that the comprehensive capabilities of the Doubao Large Model 1.5Pro version have been significantly enhanced, with low training/inference costs and an efficient model structure, greatly improving multimodal capabilities and reasoning abilities, leading globally in multiple public evaluation benchmarks. Doubao officials emphasized that no data generated by other models was used during the model training process, ensuring a solid foundation without shortcuts.
Specifically, the Doubao Large Model 1.5Pro leads globally in multiple public evaluation benchmarks, including knowledge (MMLU_PRO, GPQA), code (McEval, FullStackBench), reasoning (DROP), and Chinese (CMMLU, C-Eval).
Among them, Doubao-1.5-pro achieved better comprehensive scores in multiple benchmark tests than industry-leading models like GPT-4o and Claude 3.5 Sonnet, setting a historical best record. Meanwhile, the new Doubao visual understanding model Doubao-1.5-vision-pro leads globally in visual understanding capabilities.
Additionally, based on the Doubao 1.5 base model, breakthroughs in RL algorithms and engineering optimization have led to the development of the Doubao deep thinking model without using other model data. The phase progress Doubao-1.5-Pro-AS1-Preview has achieved industry-leading results on AIME.
Doubao-1.5-pro has been launched in grayscale on the Doubao App, performing excellently under massive requests. Developers can also directly call the API on the Volcano Engine.
Step-Video V2 Version Released
On January 22, 2024, Jiexing Star officially released the Step-Video V2 version.
According to the official introduction, compared to the Step-Video V1 released in 2024, the V2 version has more model parameters, significantly improved semantic understanding and instruction-following capabilities, and stronger generation capabilities in complex movements, aesthetic characters, visual imagination, basic text generation, native bilingual input in Chinese and English, and camera language.
It is reported that Step-Video has continuously upgraded from V1 to V2 in three aspects: the integration of VAE models, DiT architecture, and reinforcement learning, as well as the application of multimodal large models.
Step-Video V2 is now available for trial application on the Yuewen web platform.
Fonte da se uno
Dichiarazione di non responsabilità: le informazioni sopra riportate sono fornite da ifanr.com, indipendentemente da Chovm.com. Chovm.com non rilascia alcuna dichiarazione o garanzia in merito alla qualità e all'affidabilità del venditore e dei prodotti. Chovm.com declina espressamente qualsiasi responsabilità per violazioni relative al diritto d'autore dei contenuti.
